[openib-general] Re: [PATCH] SRP: don't use TX IU after freeing it

[Roland Dreier]

    Grant> Christoph is right. Even if the code works, it's risky to
    Grant> muck with the dma_addr_t contents. I'll try to look at this
    Grant> tomorrow and if I have a better idea, propose it.  (I'm
    Grant> pessimistic that I'll have a better idea though)

I don't think I buy this.  The DMA mapping API is giving us a
gather/scatter list where each entry is a dma address X and a length
Y.  The only manipulation we're doing is feeding this to the hardware
as a list of chunks at addresses X, X + PAGE_SIZE, X + 2 * PAGE_SIZE,
on up to whatever is required to cover the length Y.  This is exactly
what any bus master would do when performing DMA, so I don't see how
it is incorrect.

 - R.